Optical Breakthrough AI redefines the light perception system

In multi-scenario applications such as perfect photos, exquisite videos, and scene recognition, in addition to the help of core hardware products, algorithms are also a hero to achieve perfect results.

“Our product has encountered a problem and I want to find an algorithm company.” This is what the other party said when he visited a non-first-line camera module company earlier.

In fact, this is just the tip of the algorithm iceberg, because the module factory needs to use the algorithm to cooperate with the test, and it returns to the terminal market. In fact, the terminal manufacturer not only fights the hardware but also fights the algorithm. This shows that the brand side values ​​the algorithm. degree.

A few days ago, Jun Jun learned from the supply chain that in the single-camera era, mobile phones mainly focus on improving the image resolution in terms of taking pictures. However, in the multi-camera era, the camera becomes smarter and smarter. It reduces noise and speeds at night. Focus, optical zoom, and depth of field applications have shown good results and development prospects, and have also promoted the development of the smartphone camera market.

Of course, behind the smarter camera is the product of hardware embracing AI. In other words, the support of hardware and algorithms is the fundamental source of achieving the above effects. According to feedback from market participants, in the era of intelligent interaction, the algorithm will Become an important part of the optical breakout competition. 

However, what is the future development trend of AI algorithm companies? In this regard, Wang Junjun deliberately asked many people in the industry. Some people in the industry bluntly said that AI algorithm companies can break through the limitations of hardware imaging capabilities on the basis of existing hardware capabilities, improve imaging quality, expand imaging effects, and maintain a relatively high level of imaging. Under the premise of low power consumption, it meets the requirements of performance and user experience, greatly enriching the camera functions of smart devices, and its ultimate goal is to achieve global optimization from a single point of optimization, so as to provide consumers with a better mobile phone shooting experience.

Thinking under terminal differentiation: Hardware embraces AI as an important breakthrough tool 

With the increasingly fierce competition of smartphones, especially Android phones, with the configuration becoming more and more homogenous, manufacturers have begun to take the route of differentiation, and screens and cameras have become the main paths for differentiation.

In the field of cameras, the path to differentiation is innovation. Looking back on the development of the mobile phone camera industry in the past, it has mainly experienced seven stages: preliminary application, introduction of innovative features, high-pixel era, high-definition video era, multi-camera era, and depth camera era, reflecting the continuous upgrade and function of smartphone camera performance and functions. Innovation.

However, with the gradual failure of Moore’s Law, the time-consuming and exponential growth of R&D costs for the development of next-generation process technology chips, camera-related hardware has gradually encountered development bottlenecks. In this case, through calculations Visual algorithms and auxiliary software to strengthen the processing capabilities of the hardware, break through the performance bottleneck, and optimize the overall imaging effect of the camera have become the main development ideas for major mobile phone manufacturers to improve the performance of the imaging system.

A closer look reveals that AI algorithms have become one of the protagonists at the new mobile phone product launch. With the popularization of hardware such as multi-camera and depth cameras, the image information that can be obtained by the mobile phone photography function is more abundant. The functions enhanced by AI algorithms such as low-light photography, backlight photography, and HDR have become new competition points for major mobile phone manufacturers. AI algorithms also bring more possibilities for face interaction and somatosensory interaction for smart phones. According to data from the “2020 China Artificial Intelligence Mobile Phone White Paper” (the “White Paper”), as of 2019, the coverage rate of AI shooting algorithms in Android phones at different prices in China has reached 91.6%.

In this regard, a veteran in the industry bluntly said when communicating with Observer: “The terminal not only competes with hardware, but also competes with algorithms, because for images, algorithms are an important tool to make up for the limitations of the hardware itself.” In other words, hardware equipment determines The lower limit of the camera’s performance is determined, while the algorithm and software determine the upper limit of the camera’s capabilities.

“Terminal manufacturers pay great attention to algorithms. It can be seen that almost every model in the Android camp has AI algorithm support. Apple this year released a D TOF camera. In fact, it is to achieve interaction, and the interaction must use algorithms. In other words, not only now, but in the future, algorithms will also become just-needed products for the major terminal brand manufacturers to compete.” Another camera module industry insider said deeply. Taking Megvii as an example, the “White Paper” shows that about 75% of the high-end Android phone series released by major domestic manufacturers use Megvii AI algorithms, including AI algorithms that cooperate with Megvii.

However, as an important weapon to break through in differentiated competition, the characteristics of the AI ​​algorithm itself are very distinct: different mobile phone application scenarios require different algorithms. The above-mentioned camera module manufacturers in the industry bluntly said when communicating with Observer, “Whether it is Apple, Huawei, Xiaomi, OPPO or vivo, due to the different pursuit of image effects by consumer groups, the adjustment of algorithms is also different.”

Mobile phone imaging system development trend: single point optimization to global optimization 

The importance of the algorithm can be directly reflected from the terminal. So what is the future development trend of mobile phone imaging systems? A person from a camera module manufacturer bluntly said that the trend will be from single point optimization to global optimization. Mobile phone supply chain players are looking for ways to break through the overall effect of mobile phone shooting. Starting from its own AI technology innovation strength and practical experience, Megvii took the lead in proposing the concept of AI redefining the light perception system.

So, what exactly is the light perception system? 

According to statistics, at least 90% of the total information received by the human senses from the external world is obtained through the use of light through the visual system. However, light is not represented alone, it needs a certain carrier to realize its various characteristics; and the sensory and knowledge parts are various sensors, and the popular understanding is equivalent to the human eye and brain. 

The traditional optical perception system is that the information between the systems is only connected through the “small water pipe”, which uses the camera module, lens, light source and light source control system to collect target object data, and uses the control system, vision software and data algorithm library for image analysis And processing, software and hardware systems complement each other to give the downstream visual capabilities, but due to the independent optimization of each system, the coordination function is limited.

In the view of the observer, the traditional optical system belongs to single-point optimization, and the addition of AI makes the traditional optical system closer and closer to the global optimization. 

Specifically, the addition of AI enables objects to connect various systems through neural networks in the process of light, sense, and knowledge, so as to realize the two-way communication of more information on the light, sense, and knowledge systems. In essence, the optical perception system is a whole, and they must work in coordination. Logically, by expanding the throughput of the overall information pipeline, a more effective global optimization can be formed between different hardware devices, so that the effect of the entire system is better. Go up a step. Taking the super image quality technology as an example, Megvii has contributed to the magical marriage between computer vision AI technology and traditional photography technology. It not only achieves global optimization such as picture detail enhancement, dynamic range enhancement, and noise reduction effect upgrade, but also a special new feature. New features such as eye fusion repair, sky noise optimization, and wave ghosting optimization.

The reflection of the market is the best proof. According to the data in the “White Paper”, 10 of the top 15 mobile phones in the DXO mobile phone shooting evaluation are domestic mobile phones. Among these 10 domestic mobile phones, application or cooperation with Megvii to develop basic image quality related algorithms accounted for more than 70% . Light, sense, and knowledge are used in the entire optical market, including the camera market and the fingerprint recognition market. In the process of fingerprint recognition under the screen, global optimization will inform each link and applicable parameter information throughout the entire process, solve complex problems in various specific application scenarios, and propose practical solutions.

Optical sensing system is a high-value component of mobile phones, accounting for nearly 50%, and iteratively upgraded, it has become the focus of mobile phone manufacturers’ competition and an incremental factor that breaks through the bottleneck period of the mobile phone market.